To evaluate the reliability, validity, and sensitivity to change of tibiofemoral (TF) narrowing on lateral radiographic views.
To evaluate the reliability, validity, and sensitivity to change of tibiofemoral (TF) narrowing on lateral radiographic views.
In a natural history study of symptomatic knee osteoarthritis (OA), both lateral view and fluoroscopically positioned posteroanterior (PA) semiflexed view radiographs of the knee in 30° of flexion and with weight bearing were obtained at baseline and at 30 months. Test–retest reliability was evaluated using repeat radiographs, with joint space width measured using electronic calipers. All radiographs were scored on a 0–3 scale, and progression of joint space loss was defined as narrowing of the joint space by 1 grade. We evaluated sensitivity to change compared with the PA view. We evaluated validity by examining whether knees with progression showed expected malalignment on full-limb films.
Test–retest reliability of the TF joint space using the lateral view had a root mean square error of 0.303 mm, with 92.5% of repeats within 1 mm. More knees showed progression on the lateral view alone (n = 41) than on the PA view alone (n = 27). Compared with knees without joint space loss, knees with medial compartment loss on the lateral view only were more varus malaligned (P < 0.001), while those with lateral compartment loss were more valgus malaligned (P = 0.008).
In the assessment of TF joint space loss, lateral view radiographs are reliable, valid, and more sensitive to change than fluoroscopically positioned PA radiographs.
Assessment of change in knee cartilage over time is a critical element of clinical trials of disease-modifying agents (1–4) and of natural history studies of knee osteoarthritis (OA) (5–8). In general, anteroposterior (AP) or posteroanterior (PA) view radiographs have been used to assess the tibiofemoral (TF) joint space width (JSW) as a correlate of cartilage thickness, and studies have evaluated the loss of JSW as a proxy for cartilage loss. Both fluoroscopically positioned and nonfluoroscopic AP or PA views with the knee in flexion achieve excellent reliability in measurements of the medial joint space. With fluoroscopically positioned radiographs, investigators can detect joint space change in the medial compartment in the time frame of a clinical trial (9–11).
For fluoroscopic positioning to be advantageous, the beam must be directed precisely parallel to the upper border of a curved tibia in the medial compartment. This requires special training of radiographic technologists and considerable quality control (12), and it entails additional cost and use of higher radiation doses than conventional radiographs. An additional limitation of this method is that only the medial compartment joint space is the focus of imaging, and this may remove from consideration the 10–20% of patients with primary lateral involvement with knee OA (13). The low sensitivity to change of nonfluoroscopic methods, together with the difficulty in implementing fluoroscopic positioning methods, has led to a scarcity of practical radiography options for clinical trials and other longitudinal studies of knee OA.
To provide another option for longitudinal assessment of the joint space, we propose the use of lateral view radiographs to image the TF joint space. The positioning of the knee for this view is simple and easy to standardize, can be done without fluoroscopic positioning, and the joint space can be visualized in both TF compartments. Figure 1 shows an example of a lateral view radiograph with the JSW in the medial and lateral compartments identified. While the lateral view of the knee could be used to evaluate the TF joint space over time, there has been no evaluation of its short-term reliability or its validity for the detection of long-term joint space loss.
We used data from a 30-month natural history study of symptomatic knee OA to evaluate lateral views for test–retest reliability and validity in association with mechanical axis, a strong predictor of long-term joint space loss (14). In addition, we compared the sensitivity of the lateral view with that of a fluoroscopically positioned PA view in the detection of joint space loss over time.
Participants in this study were subjects from the Boston Osteoarthritis of the Knee Study (BOKS). BOKS was a 30-month natural history study of symptomatic knee OA. Subjects for that study were recruited from the Veterans Administration (VA) Boston Healthcare System or from the Boston-area community by newspaper advertisements. For inclusion in the study, subjects had to be at least 45 years of age with symptomatic knee OA, which was defined as pain on most days in at least 1 knee, with a definite osteophyte in the painful knee. Details on recruitment and followup in this study have been reported elsewhere (6).
Lateral view and fluoroscopically positioned PA semiflexed view radiographs were obtained from all subjects at enrollment and again after 30 months of followup. Lateral views were with weight bearing with the knee in 30° of flexion confirmed by goniometry using the Framingham Osteoarthritis Study protocol (15). This entailed obtaining 2 images per subject, 1 of the right and 1 of the left knee. Subjects stood parallel to the bucky, with the lateral joint line of the knee being radiographed touching the bucky and the foot of that knee parallel to the bucky. The front leg was imaged, and the toe of the back leg touched the heel of the front leg with the back leg locked in full extension. The x-ray technologist measured the angle of the knee from the inside (medial) of the knee. Fluoroscopically positioned PA radiographs were obtained according to the Buckland-Wright protocol (16). For evaluation of test–retest reliability, we obtained a repeat lateral radiograph from 20 subjects at enrollment, taken hours later on the same day. A full-limb radiograph was obtained from subjects at 15 months of followup. The protocols were approved by the Boston University Medical Center and VA Boston Healthcare System Institutional Review Boards. Details of the reading of PA and full-limb radiographs and the reliability of these readings have been reported elsewhere (6).
All views in lateral and fluoroscopically positioned PA radiographs were scored on a 0–3 scale. The atlas of the Osteoarthritis Research Society International (OARSI) was used for grading the PA view (17), and the lateral view was graded using an atlas developed for the Framingham Osteoarthritis Study (15). The enrollment and followup radiographs for each view were read as a pair, with the sequence known. Joint space narrowing (JSN) was defined as an increase of at least 1 grade from baseline to the 30-month followup (e.g., a change from grade 1 to grade 2). For the 20 subjects with repeated lateral radiographs at baseline, we also measured the point in the medial and lateral compartment joint space that appeared to be the narrowest visually and used electronic calipers to measure JSW at this point in each compartment (Figure 1). Mechanical malalignment was read in degrees of angulation to the varus or valgus side from the full-limb radiograph (6).
We assessed the reliability of the JSW at the narrowest point of the TF joint space based on the electronic caliper measurement. First and second radiographs were measured on different days by the same observer (DTF), with the order of films mixed the second time, and test–retest reliability was evaluated using the root mean square (RMS) error from an analysis of variance model with the subject knee as the predictor. We also used the standard suggested by Mazzuca et al (12) that knees in which the measured JSW varied by >1 mm on repeat films were unacceptable. The reliability was also graphically examined using the Bland and Altman method (18), as well as by the percentage of repeat films within 1 mm.
For each radiographic view, longitudinal joint space loss was defined as an increase of 1 or more units on the joint space scale, from the appropriate atlas, from baseline to 30 months of followup. Knees with a joint space grade of 3 at baseline (bone-on-bone) could not change on the scale, so knees evaluated as grade 3 on either the PA or lateral view at baseline were excluded from longitudinal analyses. Sensitivity to change was gauged as the percentage of the total number of knees with joint space loss of at least 1 grade detected on either the lateral or PA view. Films were read during the year of the final examination. Unknown to the primary observer (DTF), we interspersed already-read longitudinal films with unread films for the primary observer, in order to determine the reproducibility in reading joint space loss. Two or 3 previously read longitudinal films were selected randomly and inserted randomly in each batch of 20 unread films, with the observer unaware of the previous readings. The interval between initial and subsequent readings was, on average, 6 months. To assess the reliability of joint space loss readings in the lateral and fluoroscopically positioned PA views based on changes in the 0–3 grading scale, as defined above, we compared kappa statistics.
To validate progression seen on lateral radiographs, we determined the association between changes on the lateral view with mechanical malalignment, which is known to be a potent risk factor for TF joint space loss (14). Knees were classified into 2 groups according to the presence or absence of joint space loss on the lateral view (this included knees with and without joint space loss on the PA view). Our main validity analysis was the comparison of mechanical alignment between knees in these 2 groups. Since malalignment predisposes knees to joint space loss in a specific compartment (e.g., varus malalignment predisposes a knee to joint space loss in the medial compartment), this analysis was done separately for each compartment. For example, knees with joint space loss in the medial compartment were compared with knees with no loss in either compartment; malalignment was measured continuously as the degree of varus angulation, with valgus angulation characterized as a negative angle. Wilcoxon's test (19) was used to determine the significance of these differences in degree of malalignment between the groups. As a secondary analysis of validity, we eliminated knees with joint space loss on the PA view and repeated this analysis. In this secondary analysis, knees with change only on the lateral view were compared with knees with no change on either view.
At baseline, there were 323 subjects with 606 native knees that had grading PA and lateral view radiographs. We excluded 124 knees that, on either of these views, had bone-on-bone narrowing at baseline. Of these 124 knees, 99 had bone-on-bone changes on the lateral view and 68 on the PA view; 43 knees had bone-on-bone changes on both views. This left 482 knees eligible for followup. To be eligible for this analysis, subjects had to obtain repeat radiographs of the knee (same views) at the 30-month followup. Information on longitudinal changes on both views was provided by 355 knees. Further details on the demographic and clinical characteristics of these subjects have previously been reported (20).
The test–retest reliability of the narrowest point of the TF JSW on the lateral view had an RMS error of 0.303 mm. With a mean joint space of 3.03 mm, the coefficient of variation was 10%. The Bland and Altman plot shown in Figure 2 shows no systematic deviations from random scatter, which indicates that the reliability of joint space evaluation is not influenced by the amount of joint space in the compartment. The mean ± SD difference between the test–retest readings was −0.01 ± 0.43 mm, with plotted limits of agreement of −0.88 mm and 0.86 mm. Four of the plotted points 40 (10%) were outside the limits of agreement, and only 3 knees (7.5%) had repeat assessments that differed by >1 mm.
Of 355 knees without grade 3 joint space on either view at baseline, 120 (34%) showed change on either the PA or the lateral view using the 0–3 grading scale and defining progression as at least a 1-grade change (Table 1). Of these 120 knees, 51 (43%) had progression on both views, 41 (34%) had progression on only the lateral view, and 28 (23%) had progression on only the PA view. Of the 92 knees showing progression on the lateral view, 73 had progression in only the medial compartment, and 19 had progression in only the lateral compartment. If progression was defined as at least a 2-grade change in the 0–3 score, 9 knees had this degree of change on both views, 11 on the lateral view only, and 3 on the PA view only. The reliability of assessing progression was the same (κ = 0.73, P < 0.001) for PA and lateral views.
|No. of knees||No. of knees with change||% of knees with change on at least 1 view|
|Change on at least 1 view||120||0||0|
|Change on both views||0||51||43|
|Change on only the fluoroscopically positioned posteroanterior view||0||28||23|
|Change on only the lateral view||0||41||34|
Of knees with longitudinal change information and full-limb radiographs (Table 2), 213 knees with no progression had a mean mechanical axis of 1.92° of varus alignment, while for the 64 knees with medial compartment progression on the lateral view, the mean mechanical axis was 6.42° of varus alignment (P < 0.001). The 13 knees with lateral compartment progression on the lateral view had a mean mechanical axis of −2.77 (negative value is assigned to valgus alignment). Compared with knees with no progression on the lateral view, knees with lateral compartment progression came from limbs that were more malaligned in a valgus position (P < 0.001).
|Group (n)||Angulation, mean ± SD degrees†||P‡|
|Knees with medial compartment change (64)§||6.42 ± 3.79||<0.001|
|Knees with lateral compartment change (13)§||−2.77 ± 4.02||<0.001|
|Knees with no change (213)||1.92 ± 3.56|
In secondary analyses in which we excluded knees with progression on the PA view, there remained 213 knees with no progression, 28 knees with medial compartment progression, and 7 knees with lateral compartment progression on the lateral view. Knees with no progression had a mean mechanical axis of 1.92° of varus alignment; knees with medial compartment progression had a mean mechanical axis of 5.75° of varus alignment, which is significantly different (P < 0.001). The 7 knees with lateral progression had a mean mechanical axis of −2.14° of valgus alignment, which is significantly different from that of the knees with no progression (P = 0.019).
Lateral view radiographs have good reliability, better sensitivity to change than fluoroscopically positioned PA views, and are valid for the assessment of TF joint space change. Lateral radiographs provide a reasonable alternative to fluoroscopically positioned AP or PA radiographs in longitudinal studies of knee OA.
We are not aware of prior attempts to evaluate the reliability and validity of the lateral view radiograph as a method to evaluate the TF joint space. The test–retest reliability of the lateral view (RMS error) is similar to that of optimally acquired fluoroscopic AP views. Measuring the reliability of fluoroscopically positioned AP radiographs at one clinical site, Mazzuca et al (21) noted an SEM (the same measure as our RMS error) of 0.32 for those of intermediate quality (versus 0.40 for poor quality and 0.25 for high quality). From different sites, the SEM was uniformly worse than we reported for the lateral view (range 0.42–0.56). Thus, the reproducibility of joint space measurement for the TF compartments visualized from the side (the lateral view) without fluoroscopic positioning is roughly equivalent to that reported elsewhere for the fluoroscopically positioned AP view (21). We should note that, while the PA and AP fluoroscopically positioned views are accepted as methods by which to follow JSN, there is, in fact, an absence of longitudinal validation for joint space loss in the PA/AP view in other studies. However, since the AP and PA views have both been used in studies documenting the potent effect of malalignment on joint space loss (6, 14), it would be reasonable to state that AP and PA views have been validated as methods to track joint space loss.
Whereas magnetic resonance imaging (MRI) techniques offer a means of directly evaluating the volume of cartilage in a knee, the Food and Drug Administration continues to require, and trialists continue to use, radiographic progression as the primary method of assessing joint space loss. Recent recommendations for the conduct of clinical trials of disease-modifying treatments in OA still regard radiographs as the primary method of assessment (22). Since the current gold standard for trials is the fluoroscopically positioned radiograph, that was the appropriate comparator. To compare lateral and PA/AP views on their correlation with cartilage loss on MRI would require an exploration into locations of cartilage loss, since the lateral view is taken with the knee in more flexion than most AP or PA views. This complicated issue will be the topic of a subsequent study.
We used goniometry to evaluate the consistency of knee flexion and prespecified an angle to use for followup of subjects, but this could be more easily accomplished with a fixed frame or other method to ensure a certain flexion angle of the knee.
We used methods of validation that have not been previously tested for radiographs, yet we recognize that they could be questioned. Testing whether malalignment increases the likelihood of progression on the lateral view is a form of construct validity. If the lateral view were a valid method for evaluating progression, malalignment should increase the risk of progression as detected on the lateral view. While we do not know exactly what progression on only the lateral view signifies, its association with malalignment suggests that it correlates with cartilage loss, just like joint space loss on the PA view. Analyses did not adjust for correlated knees, which is challenging using Wilcoxon's test. We performed additional parametric analyses using the degree of malalignment as an outcome and progression on the lateral view as a predictor and used generalized estimating equations to adjust for interknee correlation. In this analysis, differences in alignment between groups remained statistically significant (P < 0.0001 for medial progression, P = 0.0038 for lateral progression).
Lateral radiographs were read unblinded to sequence. However, we read a subset of 21 subjects' radiographs (both knees at each of 3 visits) blinded to sequence, and the reliability of the blinded readings was high and showed no tendency toward an increase in our tendency to label a knee as having progressed. While studying vertebral fractures using serial spine radiographs, Ross et al (23) found that compared with reading blinded to sequence, reading unblinded to sequence better detected known risk factors for fracture and led to fewer errors in characterizing fractures. We contend that reading all films unblinded to sequence is a more accurate way of assessing progression and is valid, especially if the reader does not know the risk factor status of the subject.
We used a semiquantitative approach to evaluate joint space loss, and a pure quantitative approach in which JSW was measured using calipers may provide more sensitivity to change than our approach. A method would need to be developed to accommodate both medial and lateral joint space loss.
There are potential drawbacks to using the lateral view. First, more knees will need to be excluded at baseline because they show bone-on-bone narrowing. Not only are the lateral views more sensitive to change than the AP or PA views but, perhaps because they visualize knees at 30–50° of flexion, knees are often markedly narrowed on the lateral view but still show considerable remaining joint space in the same compartment on the PA/AP view. We excluded from followup 99 knees that, at baseline, had bone-on-bone changes on the lateral view and would have excluded a smaller number (68 knees) if we had relied solely on the PA view. Thus, if the lateral view were used to evaluate the progression of narrowing, the number of eligible knees would be a bit lower. Second, medial and lateral compartments are sometimes hard to differentiate on the lateral view. We believe that this presumed drawback is due to lack of familiarity with the shapes and sizes of the medial and lateral femoral condyles and tibial plateaus on the lateral view. We currently find an unreadable lateral view a rarity.
In conclusion, the lateral view of the knee in a weight-bearing position can be used with validity and reliability to evaluate the joint spaces of the medial and lateral TF compartments.